FC100 stands for Fewshot-CIFAR100, which is split from the CIFAR100 dataset and is used for few-shot learning. The FC100 dataset contains 100 categories, each with 600 images, for a total of 60,000 images.

The FC100 dataset is not divided by class, but by superclass. The dataset contains 20 superclasses (60 categories), of which the training set has 12 superclasses, the validation set has 4 superclasses (20 categories), and the test set has 4 superclasses (20 categories). The image size is 32 × 32 pixels, and the smaller image size makes small sample learning more challenging.